© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Jacobian LensNEW
    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 22-GEMMASCOPE-2-TRANSCODER-262K
    4. 44200
    Prev
    Next
    INDEX
    Explanations

    - `jwt.decode(token, ...)`- `soup.title.text`- `std::env::var("DATABASE_URL").expect(...)`- `jwt.get_unverified_header(token)`- `response.xpath('//title/text()').get()`- `let db_pool = PgPool(...)`- `const horasFaltantes = ...`- `value >>> 0` (unsigned right shift)- `std::env::var("OPENAI_API_KEY").expect(...)`- `os.environ.get("COINBASE_API_KEY")`- `os.environ.get()`The common theme is code, specifically involving:1. **Decoding/Parsing**: `decode`, `text`, `get_unverified_header`, `xpath`, `title.text`, `find`2. **Variable/Environment Access**: `expect`, `get`, `env::var`, `os.environ.get`3. **Function/Method Calls**: `()`, `(`, `("` in the TOKENS_AFTER list, and general structure in texts.4. **Specific Code Constructs**: `value >>> 0` (unsigned right shift).The neuron seems to activate on code, particularly where data is being extracted, parsed, or environment variables are being accessed. The presence of `decode`, `expect`, `get`, and code snippets like `value >>> 0` or `env::var` points towards code processing.Let's try to connect MAX_ACTIVATING_TOKENS and TOKENS_AFTER_MAX_ACTIVATING_TOKEN:- `text` is followed by `()`, `(`, `("`. Suggests function calls involving text.- `decode` is followed by `()`, `(`. Suggests function calls.- `expect` is followed by `(`. Suggests function calls or method calls.- `value` is followed by `0` and `right` in the TOKENS_AFTER_MAX_ACTIVATING_TOKEN, but also `>>> 0` in the text. This points to bitwise operations or type casting.The explanation needs to be 3-20 words, find patterns, and be specific.Consider the keywords: `decode`, `expect`, `get`, `value`.Consider the following tokens: `()`, `(`, `0`, `right`.code parsing and variable access

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_22_width_262k_l0_small_affine
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     открытия
    0.14
     oito
    0.13
     Salas
    0.13
     eight
    0.13
     zehn
    0.12
     ocho
    0.12
    덟
    0.12
    rocessing
    0.12
    alog
    0.12
    eight
    0.12
    POSITIVE LOGITS
    Ka
    0.14
     unst
    0.13
     reputation
    0.13
     ע
    0.13
     вважа
    0.13
    Yea
    0.13
     ka
    0.13
    Vij
    0.13
    ऎ
    0.12
    Feder
    0.12
    Activations Density 0.003%

    No Known Activations